ENHANCING PLC-WIFI NETWORK CONNECTIVITY AND RADIO ENVIRONMENT MAP CONSTRUCTION THROUGH FEDERATED LEARNING

Abstract

In today’s rapidly evolving industrial landscape, the intersection of technology and operational processes is increasingly evident. As industries transition into the digital era, there’s a pressing need for powerful, efficient, and secure communication infrastructures.Central to this transformation is the deployment of hybrid broadband Power Line Communication (PLC) and Wi-Fi networks, adept at catering to the diverse demands of modern digitalized industries. These networks, with their unique blend of technologies, offer promising solutions for enhancing communications within challenging indoor industrial spaces. However, determining the coverage efficiency of such networks remains a significant challenge. In addressing this challenge, the application of radio environment mapping (REM) emerges as a critical tool. REM provides quantitative metrics and graphical visualizations to assess network coverage and performance across diverse industrial infrastructures.

First, a real industrial network has been made in the University of Ulsan using PLC and Wi-Fi hybrid network. We explore potential applications of a hybrid broadband Power Line Communication (PLC) and Wi-Fi in an indoor hospital scenario. As, modern hospitals urgently require a mobile, high-capacity, secure, and cost-effective communication infrastructure. It utilizes the existing power line cables and Wi-Fi Plug-and-Play devices for indoor broadband communication. Broadband Power Line (BPL) adaptors with Wi-Fi outputs are used to build an access network in hospitals, particularly in areas where the wireless router signal is poor. The Tenda PH10 AV1000 ACWI-FI POWER LINE ADAPTER is a set of BPL adapters that offer operational bandwidth of up to 1000Mbps. These adapters are based on the HomePlug AV2 protocol and can provide a data rate up to 200 Mbps on the Physical Layer. An experiment using the PLC Wi-Fi kit is carried out to show that a Wi-Fi and PLC hybrid network is the best candidate to provide wide range of practical applications in a hospital including, but not limited to, telemedicine, electronic medical records, early-stage disease diagnosis, health management, real-time monitoring, and remote surgeries.

Furthermore, coverage prediction stands essential to optimizing radio networks, thereby enhancing user experience. A number of path loss models and advanced machine learning algorithms have been developed with high prediction performance. However, these models conventionally operate within a centralized learning paradigm. Conversely, in this article, we proposed a novel decentralized approach based on a federated learning long short-term memory (LSTM) model to predict accurate network coverage in indoor settings. The proposed FedLSTM is a method that lets multiple users, or clients, train the model without sharing their personal data directly with a central server. In an experimental setup, we utilized real data collected from numerous clients moving along different paths. The FedLSTM model is evaluated in terms of root mean square error (RMSE), mean absolute error (MAE), and R2. Moreover, compared with a centralized counterpart, FedLSTM shows a slight rise in RMSE from 2.4 dBm to 2.5 dBm and an increase in MAE from 1.7 dBm to 1.9 dBm. In addition, we evaluate the proposed FedLSTM considering variations in the number of participating clients and in the number of local training epochs. Results show that even devices with limited computational power can meaningfully contribute to training the federated model, with fewer epochs achieving competitive results. Graphic analyses of the radio environment maps (REMs) drawn from both FedLSTM and the centralized LSTM highlight their similarities. However, FedLSTM provides data privacy for clients while reducing communications overhead and server strain.